Qwen 3.5 397B-A17B — GB200 NVL72 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) on Qwen 3.5 397B-A17B. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
Throughput at 129 tok/s/user on Qwen 3.5 397B-A17B: GB200 NVL72 hits 7928 tok/s/chip, GB300 NVL72 hits 5487. Per-million costs land at $0.07 and $0.12 respectively. GB200 NVL72 is 85% cheaper per token; GB200 NVL72 delivers 44% more tok/s/chip.
GB200 NVL72 / GB300 NVL72 on Qwen 3.5 397B-A17B at 211 tok/s/user: 2512 / 1856 tok/s/chip, $0.21 / $0.35 per million tokens. GB200 NVL72 is 69% cheaper per token; GB200 NVL72 delivers 35% more tok/s/chip.
Toward the upper edge of the 48–374 tok/s/user interactivity band, at 293 tok/s/user on Qwen 3.5 397B-A17B: GB200 NVL72 runs 1208 tok/s/chip at $0.41/M tokens, GB300 NVL72 runs 759 at $0.81/M. GB200 NVL72 is 96% cheaper per token; GB200 NVL72 delivers 59% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:7928.1GB300 NVL72:5487.2 | GB200 NVL72:2512.1GB300 NVL72:1855.9 | GB200 NVL72:1207.7GB300 NVL72:759.3 |
| Cost ($/M tok) | GB200 NVL72:$0.065GB300 NVL72:$0.120 | GB200 NVL72:$0.206GB300 NVL72:$0.348 | GB200 NVL72:$0.415GB300 NVL72:$0.812 |
| tok/s/MW | GB200 NVL72:4239647GB300 NVL72:2588300 | GB200 NVL72:1343379GB300 NVL72:875409 | GB200 NVL72:645842GB300 NVL72:358168 |
| Concurrency | GB200 NVL72:~467GB300 NVL72:~140 | GB200 NVL72:~25GB300 NVL72:~17 | GB200 NVL72:~5GB300 NVL72:~3 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.